基于类内类间判据与遗传算法的故障特征选择方法

Fault feature selection method based on within-class and among-class criterion and genetic algorithm

针对原始故障数据集因"高维"和"海量"引发的"维数灾难"问题,提出一种基于类内类间距离判据和遗传算法相结合的故障特征选择方法.在提取出时域、频域、小波包频带能量作为描述系统状态的原始故障特征集基础上,经类内类间距离判据初次选择剔除不相关特征之后,引入遗传算法二次选择去除冗余特征,得到一种近似最优特征子集.结果表明:基于类内类间距离判据和遗传算法的故障特征选择方法可以剔除不相关和冗余特征,最终得到精简特征子集,并且筛选出的特征子集对故障类型的判别有很高的识别能力.

Aimed at the ＂dimension disaster＂ problem caused by ＂multi-dimension＂ and ＂massiveness＂ of faults data set, an effective fault feature selection method is proposed based on within-class and among- class distance criterion and genetic algorithm. The time-domain and frequency-domain characteristics and frequency band energy of wavelet package were extracted to make a faults data set as the description of the original system state. Then, the faults features were first selected based on within-class and among-class distance criterion and the irrelevant features were eliminated. To remove the redundant features and get optimal feature subset, the genetic algorithm was introduced to make this second selection. The result showed that the faults feature selection method based on within-class and among-class distance criterion and genetic algorithm could be used to eliminate irrelevant and redundant features and finally get the finely reduced features subset. The selected features subset would have a high identification ability of faults type.